Introduction

The lymphatic system is essential for the removal and transport of cellular waste and excess fluid from regional tissues. Series of lymphangions along lymphatic vessels pump the fluid, or lymph, proximally through lymph nodes, for immune surveillance, and ultimately returns the fluid to the venous blood stream. Cancer treatment, including lymph node removal and radiation therapy, frequently results in lymphatic insufficiency and is manifest by progressive dermal backflow, or reverse, distal flow of lymph into the dermal lymphatic capillaries.  Chronic lymphatic insufficiency can result in lymphedema, a uncurable disease marked by swelling, tissue fibrosis, and reduced immune response. Using near-infrared fluorescence (NIRF) lymphatic imaging with ICG as a contrast agent, it was recently demonstrated that dermal backflow provides an “early warning” indicator of lymphatic dysfunction, averaging eight months prior to onset of clinical symptoms of breast cancer lymphedema (PMID: 35816269). We also found dermal backflow progressed or persisted over months and years in head and neck cancer survivors (PMID: 28263428), but that it could be resolved or reduced early after cancer treatment with two weeks of physiotherapy in 75% of cases (PMID: 30694720).

Unfortunately, reliably quantifying the extent or area of dermal backflow over time remains clinically challenging owing to the limitations of planar, fluorescence lymphatic imaging devices and changing fields of view.  In this work, we develop new methods for accurate, 3-dimensional (3D) quantification of dermal backflow to provide accurate, longitudinal assessment of lymphatic dysfunction and its response to treatment. 

Methods

We integrate a stereo depth module (Intel^TM RealSense® D435i) into our NIRF lymphatic imager for concurrent acquisition of color, depth, and fluorescence images with known frames of reference. Sets of images were acquired of phantoms and of the limbs and head and neck of lifelike mannequins. Fluorescent fiducials simulated superficial lymphatic vessels and dermal backflow in humans. Image processing algorithms facilitating the extraction of 3D point clouds, the co-registration of color and NIRF images onto 3D point clouds, the stitching together of multiple point clouds, the identification and thresholding of areas of dermal backflow, and the quantitation of areas of dermal backflow of 3D surfaces are in development.

Results

The stereo depth and NIRF lymphatic imaging systems were successfully integrated and the sets of images transformed and coregistered and processed to obtain textured 3D point clouds.  Multiple 3D views were stitched together to produce more complete surface renderings of the complex head and neck surface geometries and the underlaying, superficial lymphatic networks.  We also demonstrate the use of U-Net, a deep learning technique designed primarily for image segmentation, to achieve a preliminary thresholding accuracy of 96%. To test the accuracy of the thresholding and quantification algorithms, fluorescent fiducials, mimicking dermal backflow, were affixed to the mannequin and imaged. Surface meshes were generated and the 3D areas of the fluorescent fiducials were calculated. For fiducial areas above 10 cm², decimating the point cloud to about 8% of the original number of vertices enabled rapid computation of fluorescent areas, with errors of ±3%. However, for masked areas below 10 cm², the number of vertices required for accurate area calculation increased substantially with increased errors of 6%.

Conclusion

The accurate quantification of dermal backflow on clinically relevant 3D surfaces is feasible and may provide a clinically useful measurement of lymphatic dysfunction in progressive disease and its response to treatment. Future work includes the further development and optimization of these algorithms, and their validation in a clinical study assessing the lymphatic response of early physiotherapy in head and neck cancer survivorship.